Recent advancements in both raw computing powers as well as the capabilities of cameras in recent times, it has now become possible to capture images of very high quality. This improvement however does come at the cost of the overall space required to store such high-quality images. One possible solution to this problem would be the storage of the images in low resolution and then upsampling the images to obtain the original resolutions. Despite advances in computer vision, deep learning models for super resolution have been introduced to address the challenges and thus provide promising improved performance results. This paper will explore a novel self-supervised deep learning architecture entitled Enhanced Residual Connections for Image super resolution (ERCSR) that is capable of upsampling extremely low-quality images to their higher quality. Experiments on the different data sets such as DIV2k and rice plant images are made to evaluate this model and experimental results shows that our method outperforms image enhancement. Furthermore, rice plant images are subsequently passed through disease classification layers and achieves desired accuracy for super resolution images.